Extracting a priori knowledge informing about the landscape underlying an unknown optimization problem has been proved extremely useful for different purposes, such as designing finely-tuned algorithms and automated solving techniques. Focusing on continuous domains, substantial progress has been achieved with the development of the so-called exploratory landscape analysis (ELA) approach, which provides a unified methodology for integrating features into sophisticated machine learning techniques. In particular, much efforts have been devoted to the systematic design of algorithm selection models aiming at improving existing state-of-art solvers. Nonetheless, designing the ELA features themselves is a bottleneck that can prevent further advances. The contribution of this paper is thereby two fold. Firstly, we consider the design of insightful features on the basis of the search tree constructed by the so-called SOO global optimizer, which is shown to imply an informative sampling of the search space using a limited budget. Secondly, we provide empirical evidence on the relevance of the proposed features and their potential in complementing existing ELA features for both predicting high-level problem properties, and selecting algorithms from a portfolio of available solvers. Our empirical findings are based on a comprehensive analysis using the diverse set of BBOB functions and solvers from the COCO platform.